Balance between object and background: Object-enhanced features for scene image classification

An unsupervised object-enhanced feature generation mechanism is proposed, which balances the different effects of object regions and background regions for scene image classification. The proposed method strengthens the characteristics of the whole image with object regions in a biased way and accords with the perception mechanism of humans. Furthermore, it can be easily embedded in the extraction process of existing prominent histogram-based feature representations, such as BOW (bag-of-visual-words) and HOG (histogram of orientations gradients). The paper takes BOW feature as the primary example, and presents a feature named OE-BOW. The overall results of the proposed method show an increase in the classification accuracy of about 1.0–2.5% compared to the original feature on some popular scene datasets. And the performance of the proposed method is also shown to be comparable to the recently reported results.